Zayneb C.

Key Skills

Software

Other

Top Subject Matter

AI model training and RLHF-style evaluation (prompt/response refinement)

Prompt Review Writing

AI Training Data Quality Analyst

Top Data Types

Document

Text

Image

Top Task Types

Question Answering

Prompt + Response Writing (SFT)

Entity (NER) Classification

Transcription

Freelancer Overview

AI Model Trainer (Feather), OpenAI — Project Vox. Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Other. Education includes Bachelor of Arts, Columbia University (2024). AI-training focus includes data types such as Computer Code, Programming, and Document and labeling workflows including Evaluation, Rating, and Question Answering.

Labeling Experience

Advanced AI Extern - Pfizer

TextTranscription

Built document processing pipelines in Python using layout-aware OCR to extract structured data from unstructured pharmaceutical documents. Developed retrieval-augmented generation (RAG) systems with open-source LLMs and vector search for enterprise document Q&A. Shipped an AI chat interface enabling natural-language queries over large document corpora while applying end-to-end engineering and data modeling skills. • Implemented layout-aware OCR and structured data extraction pipelines • Developed RAG systems with vector retrieval for document question answering • Built and integrated an AI chat interface for enterprise use • Worked with Python-based data processing and LLM application architectures

2026 - Present

AI Model Trainer (Feather) - OpenAI

Entity (NER) Classification

Trained OpenAI’s Codex model for code-focused performance improvements through prompt design and structured evaluation for Project Vox. Developed adversarial and edge-case prompt sets to stress-test model behavior and reliability. Provided actionable feedback to enhance code generation, reasoning, and instruction-following quality while applying strong experimental and QA discipline. • Crafted targeted prompts and evaluation protocols for Codex • Designed adversarial/edge-case test suites to assess robustness • Analyzed model outputs and delivered structured improvement feedback • Supported reliability-focused iteration for agent/code use scenarios

2026 - Present

Agentic AI & ML Intern, MIT-Incubated Stealth Startup

DocumentPrompt + Response Writing (SFT)

Engineered an agentic AI/ML system by building an LLM-driven backend for document ingestion and semantic search. Implemented orchestration using LangChain for multi-step LLM reasoning across enterprise data. Designed agentic workflows and tool-use logic for real-time financial data processing and decision support. • FastAPI backend for ingestion and chunking • Semantic search for context retrieval • LangChain orchestration for multi-step reasoning • Design of agent workflows and tool-use logic

2026 - Present

Advanced AI Extern, Pfizer — Remote Project

OtherDocumentQuestion Answering

Built document processing workflows to extract structured information from unstructured pharmaceutical documents. Developed RAG systems using open-source LLMs and vector retrieval to support enterprise document question answering. Shipped a natural-language AI chat interface over large document corpora. • Layout-aware OCR for document parsing • Structured data extraction from unstructured sources • RAG pipeline development for retrieval-backed answers • Deployment of chat/Q&A over document corpora

2026 - Present

AI Model Trainer (Feather), OpenAI — Project Vox

Worked as an AI model trainer by crafting targeted prompts and evaluating model responses for a code-focused language model. Designed adversarial and edge-case prompt sets to stress-test behavior and reliability. Provided structured feedback to improve code generation, reasoning quality, and instruction-following. • Prompt engineering for targeted model behaviors • Adversarial and edge-case test set creation • Structured response evaluation and feedback • Reliability and accuracy improvement for code generation

2026 - Present

Education

Work History